Quantum Amplitude Estimation
A quantum routine that estimates the probability of a marked outcome quadratically faster than classical sampling.
The estimation problem
Suppose a unitary A prepares a state A|0> = sqrt(1-a)|psi_0>|0> + sqrt(a)|psi_1>|1>, where the last qubit flags a marked subspace. The number a in [0,1] is the probability of measuring the flag qubit in state 1. Classical Monte Carlo estimates a by repeated sampling, achieving additive error epsilon with O(1/epsilon^2) shots. Quantum amplitude estimation (QAE) reaches the same error with O(1/epsilon) uses of A, a quadratic speedup.
How it works
QAE combines amplitude amplification with quantum phase estimation. Define the Grover-like operator Q = A S_0 A^dagger S_chi, where S_chi flips the phase of the marked subspace and S_0 flips the phase of |0>. In the two-dimensional subspace spanned by the marked and unmarked components, Q is a rotation by angle 2*theta, where a = sin^2(theta). Phase estimation on Q reads out theta, and hence a.
The eigenvalues of Q are exp(+/- 2 i theta). Running phase estimation with m ancilla qubits resolves theta to precision ~2^-m, so a is recovered to accuracy scaling like 1/M where M = 2^m is the number of applications of Q.
Why the speedup
- Classical error shrinks as 1/sqrt(N) with N samples.
- Quantum error shrinks as 1/M with M coherent applications of Q.
- The Heisenberg-limited 1/M scaling is the source of the quadratic advantage.
Practical notes
Canonical QAE requires deep coherent circuits and a quantum Fourier transform on the ancilla register, which is demanding for near-term hardware. Variants that avoid phase estimation (see amplitude estimation without phase estimation) trade some elegance for shallower circuits. QAE is the engine behind quantum speedups for Monte Carlo integration and expectation-value estimation.
In a fusion context, the same expectation-value structure appears when estimating averaged observables from simulated plasma ensembles; the machines themselves (Hyperion breeder, Aegis and MetroVolt burners) are design and simulation studies, so any such use is computational, not experimental.